search

anon998 / simple-proxy-for-tavern

GNU Affero General Public License v3.0
110 stars 6 forks source link

Any model repeats inferences after a couple of replies #1

Closed Talref closed 1 year ago

Talref commented 1 year ago

I've tried a couple of 13B models, loaded in 4bit (vicuna, gptxalpaca and supercot). After about 5 or 6 creative and fine messages I start to get repetitions of the last inference, independently of the last message sent. Backend: KoboldAI (Occam's 4bit fork) Frontend: SillyTavern

image

Talref commented 1 year ago

Seems to be a model/kobold problem, not a proxy one.